What we were sitting on
Years of building software had left us with something we had never thought of as an asset: the full, continuous record of how our company operated. Conversations, decisions, code, documents, meetings, customer interactions, across more than eight systems, accumulated over years of real work. Terabytes of it.
Like every business, our accounts valued it at zero.
How the deal happened
The opportunity came through a wind-down process: a specialist intermediary connected companies holding data with AI labs buying it. The lab's interest was specific: not documents, not files, but the connected history of how a real organisation communicates, decides and executes over time. That is the signal modern AI needs to learn how work actually happens, and it does not exist on the public internet.
The process, step by step
- Licence first. A written data licence with warranties on authority and chain of title, before a byte moved. The buyer's lawyers cared about rights before they cared about content.
- Extraction. Read-only extraction across every system: chat, email, project tools, code, documents, meetings, analytics. This took months the first time. It is the part we later turned into technology that runs in days.
- De-identification. Every person, customer and counterparty irreversibly removed: 60+ categories of identifiers, with an audit ledger recording every change. Third-party and copyrighted material filtered out entirely.
- Structuring. Raw exports are worthless to a lab. The data had to be rebuilt into time-ordered, connected records to the buyer's exact specification. This is where most of the value is created, and where most sellers would fail.
- Verification. The lab tested the dataset against its acceptance standards: provenance, privacy, structure, quality. It passed.
- Settlement. Delivery was accepted and payment was made. Terabytes in, trillions of structured training tokens out, and a wire transfer to close it.
What it taught us
The gap is the opportunity. The distance between "we have a lot of data" and "a dataset a frontier lab will pay for" is enormous, and almost no business can cross it alone. Rights, privacy, structure and buyer access each kill deals on their own.
The hard parts are repeatable. Everything that took us months, extraction, de-identification, structuring, is now software. What was a one-off windfall for us became a capability we can run for any business.
The buyers want more. The labs cannot sit with every business and do this work themselves. They told us as much. That gap between their demand and the world's unprepared supply is the business you are reading about.
We are the only advisor in this market that has been the client: same transaction, same scrutiny, our own data on the table.
The buyer and intermediary are not named here under the confidentiality customary in these transactions. Specifics are available in conversation where appropriate.